Haptic Rendering using Simplification

Comp259Sung-Eui Yoon

Overview Continuous-Adaptive Haptic

Rendering Sensation Preserving Simplification

Haptic Rendering 3 major steps

Initializing the haptic device and transferring the dataset

Collision detection between virtual objects and the probe

Estimating force• The force is fed to the generic probe

It require high update rates (1000Hz)

Haptic Rendering Process

Methods to reduce model complexity Spatial subdivision (octree, ..)

We may lose meaningful cell that can affect user.

User can perceive incorrect force when moving fast.

Static LOD Switching LOD may lead to noticeable

changes. There may be only one LOD for large model

Continuous-Adaptive Haptic Rendering It gives various level of detail at

different regions of the surface Also, reduce complexity of model

Doesn’t send whole geometry. Instead, send high resolution near the

probe. Based on View-Dependence Tree for

view-dependent rendering.

View-Dependent Simplification At preprocessing, calculate sequences of

edge collapses by model simplification method

From this, we can make a vertex hierarchy, which is represent the way how to simplify a model at run-time

View-Dependent Simplification

Switch Value• Quadric error between original geometry and

simplified one.• At run-time, we calculate projection error from this.

Dependency Information• Neighboring faces when performing collapse and

split to prevent foldover.

View-dependent Rendering Process active vertices list, which represent

current LOD of model Initialize active vertex node list with root nodes.

Reconstruction of a real-time adaptive mesh Need active triangle list There are frame-to-frame coherence

Result Image

Selecting LOD Assumption

Geometry close to probe has a higher probability of collision with the probe.

So, we need more higher resolution near the probe.

How to define appropriate resolution Bell-shaped filter, mapping table

between distance and switch value.

Run-time Algorithm Scan node of vertex list Compute the distance from the probe Determine switch value Compare this with the one stored in node

Split node if computed value is less than one in node and node satisfy dependency

Merge node with sibling if computed value is greater stored one of parent and the node meet dependency.

Optimizations Haptic and graphics buffers are updated

in an incremental fashion

The graphics and haptic rendering require different update rate 20Hz for graphics rendering 1000Hz for haptic rendering update geometry at 20 Hz

Result Use the GHOST API library.

It fails when it is pushed to run at less than 1000Hz.

Limitation Doesn’t present error metric for haptic

rendering Just use switch value for projection error.

Isn’t clear to integrate view-dependent simplification with other acceleration (Bounding Volume Hierarchy) technique for collision detection.

Sensation PreservingSimplification Key observation

Human haptic perception of geometric feature depend on the ratio between the contact area and the size of the feature

In visual rendering Consider surface deviation and the viewing

distance In haptic rendering

Contact surface area and the resolution of the simplified model

Design Issues Design multiresolution hierarchy

that : Minimize perceptible surface deviation

• Filtering the detail at appropriate resolution Reduce the polygonal complexity of low

resolution representations• Incorporating mesh decimation

Are themselves BVH of convex hull• The system take advantage of BVH of

convex hull

Definition of Resolution (Sampling) Resolution r

1D example: The inverse of the distance between two subsequent samples.

2D : the sampling resolution of an edge (v1, v2) of the mesh M at resolution, rj ,Mj

• can be estimated as the inverse of the projected length of the edge onto a low resolution representation of the mesh, Mi

Filtered Edge Collapse Two goals in the construction of

hierarchy. Generate the hierarchy with low polygonal

complexity at low resolution Filter details as we compute low resolution

These are achieved by merging downsampling and filtering operation

Convexity Constraints A surface convex decomposition for

collision detection must meet these constraints All the interior edge of a convex patch must

themselves be convex. No vertex in a convex patch may be visible

from any face except the ones incident on it The virtual face added to complete the convex

hull cannot intersect the mesh

Local Convexity Constraints

Global Convexity Constraints Too complicated to be incorporated

into filtering process Verified after the filtering

use bisection search between v3 and v3 if v3 meet the constraint

^

^

Multiresolution Hierarchy Generation Starting by computing an initial convex

decomposition and resolution for all the edges.

Edges are inserted in a priority queue with validity and resolution as 1st and 2nd keys for sorting.

Generating new LOD every time the number of convex pieces are halved. Combine neighboring convex pieces as long

as they represent a single valid convex patch.

Contact computation for Haptics Based on a penalty-based force

computation Force displayed is proportional to the

penetration depth. Bounding Volume Test Tree (BVTT)

Perform contact query as descending BVTT, which is dynamically constructed.

Generalized front tracking to exploit temporal coherence.

BVTT and generalized front tracking

Sensation Preserving Selective Refinement Only refine the lower node of BVTT if the

missing detail is perceptible. Perceptibility

Depends on magnitude of surface feature and contact area

Results

Reference M. Otaduy and M. Lin, “Sensation Preserving Simplification

for Haptic Rendering”, to be appeared in SIGGRPH2003 J. El-Sana, and A. Varshnewy, “Continuouly-Adaptive

Haptic Rendering”, Virtual Environments 2000 J.El Sana and A. Varshney. Generalized view-dependent

simplification, In Proceeding EUROGRAPHICS99, pages 83-94, 1999

M. Garland and P. Heckbert, “Surface simplification using quadric error metrics”. In Proceedings of SIGGRAPH ’97(Los Angeles, CA), pages 209 – 216. ACM SIGGRAPH, ACM Press, August 1997.

H. Hoppe, Progressive meshes, In Proceedings SIGGRAPH 96, pages 99-108. ACM SGIGGRAPH 1996